Tcl Source Code

Interestingly while open/read/stat etc don't work  file attributes does.

    P% file attributes $f
    -  -longname {NEXT SAT. 2_15_ FADED LOVERS TOUR in PROVIDENCE! 💕💞💕.eml} -shortname NEXTSA~1.EML
    P% file exists NEXTSA~1.EML
    -  1

You also can't glob for an exact match: glob $f
Dos device pathnames don't help either:
    P% file exists //?/[pwd]/$f
    -  0

(note a full path is required for the //?/ syntax)

Everything works fine if you use the shortname - but it seems if that's the case maybe the windows file code should automatically try the shortname in these cases.

This is close, but *slightly* different to the behaviour you get when using what is considered an illegal path in windows. e.g a filename ending in . or a space.

These ones do work simply by using the //?/ syntax - but.. file attributes doesn't return the shortname info.

    P% file exists [pwd]/testetc123aaa.
    -  0
    P% file exists //?/[pwd]/testetc123aaa.
    -  1
    P% file attributes [pwd]/testetc123aaa.
    could not read "C:/test/testetc123aaa.": no such file or directory
    p% file attributes //?/[pwd]/testetc123aaa.
    -  -archive 1 -hidden 0 -readonly 0 -system 0

Exact globs for trailing dot/space files also doesn't work.

It's obviously more important that the emoji names work than trailing dots or spaces which is a separate issue - I just thought the comparison with the //?/ workaround not working was interesting.

In tcl8.7 however - this with emojis works without issue that I can see.

In reference to Julian's comment regarding "file attributes", that command fails with "no such file or directory" on my Windows 10 system.  However, I can confirm that "file exists" with the short filename does return 1.

I apologize for adding my comments here instead of to Julian's but I couldn't find a way of doing it (little experience with fossil).

If it works on TCL 9.0, I am quite happy. That is anyway the aim. Emoji were always broken on TCL 8.6 and will never be really fixed.

Many people happily use TCL 8.7 or even 9.0 where all this works. Or Undroidwish, where all this works to.

Take care, Harald

The problem is here. The function MultiByteToWideChar() converts UTF-8 to UTF-16, but - in this case - the filename is not in UTF-8, it's CESU-8, because that's the encoding used internally in Tcl 8.6.

In Tcl 8.7 and 9.0, Tcl uses (modified) UTF-8 in stead of CESU-8, that explains why it works there. Possible solution, convert from CESU-8 to UTF-8 (Tcl_UtfToExternal) first.

Proposed fix [e22d09505b1c2338|here]

Remember that Tcl 8.6 uses CESU-8 as internal format, which cannot be fead directly to MultiByteToWideChar(). Tcl 8.7 and 9.0 use (modified) UTF-8 internally, that's why this 'fix' is for 8.6 only.

>Emoji were always broken on TCL 8.6 and will never be really fixed ;-)

Jan, really great work! It works like a charme! I only tried with tclsh (did not build tk), where the console output of emojies do not work. But the glob to file relation works.

I was not able to test with MS-VC6, due to [a1f11d96b8]. So, branch [e22d09505b1c2338] was tested with MS-VS2015 32 bit.

The test suite is still running. I will report here, if there are any issues.

IMHO, this may be merged.

Take care, Harald

Running the test suite caused the file related tests in ticket [26fd1451b8] to fail. But the also fail in core-8-6-branch. Due to that, it is reported in a separate ticket.

Harald

Fixed [87eae2f4507bc5cc|here]

Closing